HRV and EEG Signal Features for Computer-Aided Detection of Sleep Apnea

Abstract

Sleep apnea is a disorder in which individuals stop breathing during their sleep. Sleep apnea is categorized as obstructive, central or mixed. As an attempt to advance obstructive sleep apnea treatment, in recent years new techniques for sleep stage classification have been developed by biomedical engineers and clinicians for sensitive and timely detection of sleep disorders. In this paper, we present a compendium of features extracted from polysomnographic data acquired from twenty five patients (21 males and 4 females) suffering from sleep apnea (age: 50 ± 10 years, range 28-68 years). Sleep data were available online from the Physionet database. Time and frequency domain algorithms were applied to 3 biopotentials to extract features as follows: EEG (Hjorth Parameters, Harmonic Hjorth Parameters, Itakura Distance, Detrended Fluctuation Analysis, Relative Energy Band Percent, and Correlation Dimension), EMG (Energy Content), and EOG (Energy Content Band). Heart Rate Variability (HRV) signals were then derived from ECG signals using an Enhanced Hilbert Transform algorithm. Features extracted from the HRV signals were: R-R statistics (mean, standard deviation, maximum and minimum R-R values), detrended fluctuation analysis parameters, frequency components (LF, HF and LF/HF ratio) and approximate entropy. Results show that trends detected by these features could distinguish between different sleep stages at a highly significant level (p<0.01). These features could prove very helpful in computer-aided detection of sleep apnea.